Minireview

Two-dimensional crystallization of membrane proteins:the lipid layer strategy

Daniel Levy, Mohamed Chami, Jean-Louis Rigaud*Institut Curie, Section de Recherche, UMR-CNRS 168 and LRC-CEA 8, 11 Rue Pierre et Marie Curie, 75231 Paris Cedex 05, France

Received 11 July 2001; accepted 23 July 2001

First published online 7 August 2001

Edited by Andreas Engel and Giorgio Semenza

Abstract Due to the difficulty to crystallize membrane proteins,there is a considerable interest to intensify research topics aimedat developing new methods of crystallization. In this context, thelipid layer crystallization at the air/water interface, used so farfor soluble proteins, has been recently adapted successfully toproduce two-dimensional (2D) crystals of membrane proteins,amenable to structural analysis by electron crystallography.Besides to represent a new alternative strategy, this approachgains the advantage to decrease significantly the amount ofmaterial needed in incubation trials, thus opening the field ofcrystallization to those membrane proteins difficult to surexpressand/or purify. The systematic studies that have been performedon different classes of membrane proteins are reviewed and thephysico-chemical processes that lead to the production of 2Dcrystals are addressed. The different drawbacks, advantages andperspectives of this new strategy for providing structuralinformation on membrane proteins are discussed. ß 2001 Pub-lished by Elsevier Science B.V. on behalf of the Federation ofEuropean Biochemical Societies.

Key words: Two-dimensional crystallization;Membrane protein; Lipid layer; Lipid ligand

1. Introduction

Genomic studies of prokaryotic and eukaryotic organismsdemonstrated that membrane proteins represent about 25% ofthe predicted proteins [1]. These results emphasize the impor-tance of membrane proteins in many biological processes es-sential for life. However, compared to other protein classes,their functions are poorly understood due to the lack of struc-tural information which precludes structure^function relation-ships to be established at a molecular level. For instance, lessthan 30 original structures have been solved to atomic reso-lution, a number which lags far behind that of soluble pro-teins, with several thousands accurately determined structures.

Three main bottlenecks may explain the di¤culties in get-ting structural information on membrane proteins. First,membrane proteins are di¤cult to overexpress and only few

membrane proteins are available in a large amount enough forstructural analysis. Expression levels in mammalian, insectcells, oocytes, yeast or Escherichia coli still need to be im-proved for routine crystallization trials [2]. Production of pro-tein in non-functional form as inclusion bodies and subse-quent refolding may be a possible route towards realizingthis goal [3,4]. Second, solubilization and puri¢cation of mem-brane proteins necessitate the use of detergents and require adi¤cult and empirical biochemistry. Although the addition ofpolyhistidine motifs by gene modi¢cation that speci¢callybind nickel-chelated nitriloacetic acid (Ni2-NTA) resins hasgreatly improved the puri¢cation step, it remains that thepurity, the ¢nal concentration, the monodispersity, and thestability of the puri¢ed protein preparations are still limitingfactors in further crystallization trials. Third, amphiphilicmembrane proteins are notoriously di¤cult to crystallizeand therefore the production of three-dimensional (3D) ortwo-dimensional (2D) crystals remains one of the major chal-lenges to get structural information.

While conventional 3D crystallization has provided the ma-jority of atomic structures of membrane proteins, success ingrowing 3D crystals in detergent micelles remains relativelyinfrequent [5]. This drawback is mainly related to the di¤cultyin maintaining a crystal lattice through the sole interactionsbetween the hydrophilic domains of the proteins, the hydro-phobic domains being shielded by the detergent micelles. Tohelp with this problem, extensive works have been performedto develop new strategies. In particular, a novel approach hasbeen introduced by Michel's group in which monoclonal anti-body fragments were speci¢cally bound to proteins, increasingthe interactions between the hydrophilic domains of mem-brane proteins [6]. More recently, 3D curved membraneshave been devised as crystallization matrices with the ideathat proteins in such membrane mimetic systems should bemore stable than in micelles. Highly viscous, bi-continuouslipidic cubic phases have been successfully used as matricesfor nucleation and further 3D crystallization of di¡erent pro-teins [7,8].

As a viable alternative to the 3D crystallization of mem-brane proteins in detergent micelles, the reconstitution ofmembrane protein into lipidic membranes to form crystalscon¢ned to two dimensions has opened a new way to solvestructures by electron crystallography. If only three membraneprotein structures have been resolved to atomic resolution [9^11], more than 20 membrane proteins have been resolved tomedium resolution (4^8 Aî ) in 3D maps or projections in

0014-5793 / 01 / $20.00 ß 2001 Published by Elsevier Science B.V. on behalf of the Federation of European Biochemical Societies.PII: S 0 0 1 4 - 5 7 9 3 ( 0 1 ) 0 2 7 4 8 - X

*Corresponding author. Fax: (33)-1-40 51 06 36.E-mail address: rigaud@curie.fr (J.-L. Rigaud).

Abbreviations: Ni2-NTA, nickel-chelated nitriloacetic acid; 2D, two-dimensional; 3D, three-dimensional; cmc, critical micellar concentra-tion; His, histidine; RLM, re£ected light microscopy

FEBS 25156 23-8-01

FEBS 25156 FEBS Letters 504 (2001) 187^193

plane. Although of medium resolution, these 2D crystals haverevealed the secondary structures of di¡erent membrane pro-teins not yet crystallized in 3D. Furthermore, they allowed toanalyze structural similarities within protein families [12] orlarge conformational changes induced by di¡erent substrates[13]. However, as for 3D crystallization, the production ofwell ordered 2D crystals remains rare and many proteins con-tinue to resist e¡orts.

Except for few examples of 2D crystals of membrane pro-teins in native membranes, 2D crystals are usually obtained bythe general method of detergent-mediated reconstitution ofmembrane protein [14]. The principle involves detergent re-moval from micellar detergent solutions containing the puri-¢ed protein and a suitable combination of lipids. In additionto the large number of parameters that critically a¡ect theproduction of 2D crystals (e.g. temperature, bu¡er composi-tion, lipid to protein ratios, nature of the ternary micellecomponents), detergent removal is a key step since it controlsthe micelle to bilayer phase transition, the protein incorpora-tion into the bilayer and its crystallization. A number of ex-cellent reviews are available concerning the di¡erent methodsto produce 2D crystals and the physico-chemical mechanismsleading to their formations [15^20].

In the present paper we will focus on a new experimentalapproach for 2D crystallization of membrane proteins thathave allowed promising perspectives to be foreseen. By com-bining the concepts of surface recognition and 2D crystalliza-tion of membrane proteins by detergent removal, it has beenpossible to crystallize membrane proteins previously bound toa lipid layer spread at the air/water interface. As a main result,such a new strategy has been demonstrated e¤cient for crys-tallization of radically di¡erent membrane proteins [21]. Be-sides to represent an alternative strategy for 2D crystalliza-tion, this new approach gains the advantage to decreasesigni¢cantly the amount of material needed in incubation tri-als, opening the ¢eld of crystallization to those membraneproteins di¤cult to surexpress and/or purify.

2. Principle Of 2D crystallization on lipid layer

The principle of the lipid layer 2D crystallization method,initially developed by the group of Kornberg, is based on thespeci¢c interaction between a soluble protein and a lipid li-gand inserted in a planar lipid ¢lm at an air/water interface[22^25]. This method has been applied to more than 20 solu-ble proteins leading, for some of them, to important structuralinformation to 3 Aî resolution [26,27]. Protein crystallization

has been described to proceed in three steps: (1) molecularrecognition between soluble proteins and speci¢c lipid li-gands; (2) di¡usion of lipid^protein complexes in the planeof the ¢lm; (3) self-organization of the proteins into 2D crys-tals. It is important to stress that one important advantage ofthis method, related to the binding step and the concentrationof the protein at the surface, is that only a small amount ofprotein is required, less than 1 Wg per incubation trial.

The question as to whether this strategy was applicable tomembrane proteins was an open challenge. Indeed, its use formembrane proteins was confronted to the indispensable pres-ence of detergent with the protein, which was expected to alterthe integrity of the lipid ¢lm. However, through careful stud-ies, the feasibility of this approach to crystallize membraneproteins was recently demonstrated. 2D crystals of radicallydi¡erent membrane proteins have been produced includingTF0F1, the ATP synthase from thermophilic bacteria PS3,FhuA, a highly speci¢c porin from E. coli [21], and, morerecently, bacteriorhodopsin and the ADP/ATP transporterAnc2 from yeast mitochondria (Chami et al., in preparation)as well as aquaporin Aqp1 (S. Scheuring, personal communi-cation) and a plant H ATPase [28].

Comparative and systematic studies allowed to identify thebasic principles controlling the formation of membrane pro-tein 2D crystals on lipid layers and a working model has beenproposed. Such a model includes three steps (Fig. 1): (1) bind-ing of protein micelles to a surface lipid layer through speci¢cor electrostatic interactions; (2) reconstitution by detergentremoval of a lipid bilayer around the previously bound pro-tein micelles ; (3) further 2D crystallization in the reconstitutedlipid bilayer.

3. The three steps for membrane protein crystallization onlipid layer

3.1. Binding of the protein micelles to the lipidic surfaceOne of the main di¤culties expected in the development of

the lipid layer to crystallize membrane proteins was the pres-ence of the detergent during the binding step. During thisstep, the detergent has to be at a concentration above itscritical micellar concentration (cmc) to maintain the proteinamphiphilicity and may interfere, up to complete solubiliza-tion, with the lipid layer spread at the surface. Indeed, spread-ing a lipid layer directly on a detergent-containing dropletproved to be unsuccessful due to an immediate solubilizationof the surface lipid layer.

This drawback has been circumvented: (i) by adapting the

Fig. 1. Mechanism for 2D crystallization of membrane proteins on lipid layer at the air/water interface. (a) Binding of protein/lipid/detergentmicelles to the functionalized lipid layer spread at the air/water interface. (b) Reconstitution of proteins into a lipid bilayer upon detergent re-moval. (c) 2D crystallization.

FEBS 25156 23-8-01

D. Levy et al./FEBS Letters 504 (2001) 187^193188

order of lipid and protein additions: the surface lipid layerhas ¢rst to be formed on a drop of a detergent-free bu¡ersolution, followed by injection in the sub-phase of the proteinin a micellar form; (ii) by spreading an amount of lipids inslight excess (10^20%) to that needed for a single monolayer.This allows a maximal compression state of lipid at the inter-face, slowing down the penetration of the detergent into thepre-formed lipid monolayer. Under these conditions, when adetergent was injected at a concentration well above its cmc, agradual solubilization of the spread lipid layer was shown tooccur, but in more than 4 h, a process signi¢cantly longerthan the solubilization process of lipids in bulk [14,29,30].Interestingly, when the same amount of detergent was injectedbut with the protein, the lipid layer was extremely stable andno solubilization was observed, even after 1 month incubationat 4³C. Although a comprehensive explanation of this unex-pected stability remains speculative, it is likely related to thebinding of the protein which leads to a high concentration ofprotein micelles at the interface. This high micelle concentra-tion could induce, as already observed in many lipid^deter-gent micellar solutions [31^33], speci¢c viscous and even `gel-like' phases which in turn would slow down detergent di¡u-sion in the lipid layer. Such an interpretation is in agreementwith the observed appearance of a frozen surface upon bind-ing of the protein^detergent micelles to the lipid layer surface(see below).

Under the experimental conditions de¢ned above for thebinding step, it has been possible to bind several membraneprotein micelles to di¡erent functionalized lipid layers. Thestructures formed on the interfacial ¢lm were su¤ciently sta-ble to be transferred to carbon grids and analyzed by trans-mission electron microscopy. They appeared as thin planardomains which can be several tens of Wm in size (Figs. 2a,band 3a). At high magni¢cation, a protein-like grain wasclearly visible in these domains, denoting the e¤cient bindingof the proteins. When the proteins were large enough to bevisualized, the bound micelles appeared densely packed (Fig.3b) and, in few cases, optical di¡ractions showed hexagonalpatterns typical of a highly close packing (Fig. 2c). However,longer incubations, up to several weeks, never led to the for-mation of crystalline lattices of these protein micelles boundto the lipid layer.

A point to be stressed is that the e¤ciency of the bindingstep was found to be speci¢c for detergents with low cmcvalues such as dodecylmaltoside, Triton X-100 or other poly-oxyethyleneglycol detergents. Use of high cmc detergents, in-cluding octylglucoside, or bile salt derivatives, always faileddue to a rapid solubilization of the lipid layer. Whatever, forthose membrane proteins puri¢ed in high cmc detergents, thedi¤culty could be overcome: (1) by exchanging the high cmcdetergent for a low cmc detergent using suitable chromato-graphic columns; (2) by simply diluting the protein in a lowcmc detergent during the injection step; or (3) by using a newtype of lipid layer made of £uorinated chains. Such £uori-nated lipids, which increase drastically the hydrophobicity ofthe lipid layer, should prevent its solubilization whatever thenature of the detergent injected in the sub-phase [28].

3.2. Reconstitution of pre-bound proteins into a lipid bilayerAs stated above, 2D crystallization of protein micelles

bound to the lipid layer has never been observed. This isprobably related to the di¤culty to create hydrophilic con-

tacts between membrane proteins embedded into micelles asreported for 3D crystallization trials. Thus, to favor protein^protein contacts, detergent has to be removed, with the idea toreconstitute a lipid bilayer around the previously bound pro-tein, in order to maintain its amphiphilicity. Detergent remov-al was performed, in about 4 h, through successive additionsof Bio-Beads SM2 into the solution below the lipid layer[34,35].

It has been shown that the reconstitution of a lipid bilayeraround the pre-bound protein was strictly related to the pres-ence of lipids in the initial micellar protein solutions injectedin the sub-phase. When lipids are added to the solubilizedprotein before injection, large bilayered membranes could bereconstituted while, in the absence of added lipids, only ag-gregated structures were found at the surface after detergentremoval. Thus detergent removal allows membranes to bereconstituted from the lipids present in the ternary lipid/pro-tein/detergent micelles bound to the functionalized surface. Atthis point, it has to be noted that although reconstitution ofthe bilayer is independent of the nature of the lipid added tothe initial micelles, the lipid speci¢city is crucial for the fol-lowing 2D crystallization step.

Electron microscopy analysis of the transferred protein^lip-id ¢lms indicated that the domains formed after the binding of

Fig. 2. 2D crystallization of His-FhuA on Ni-NTA lipid layer. Ex-perimental conditions as in [21]. RLM experiments have been per-formed with a Leica DMR microscope equipped with a long dis-tance objective (U20, 0.40 nominal aperture). Images of the thesurface of the wells were recorded without any labelling by a non-intensi¢ed CCD camera (COHU 5100) without further electronicimprovement to increase the contrast. Upper part: binding step.(a) Surface of the incubation well observed by RLM showing theprotein micellar domains (light gray) formed after protein binding.Bar: 50 Wm. (b) High magni¢cation electron micrograph of thetransferred micellar domains. Bar: 200 nm. Inset: computed di¡rac-tion pattern showing the close packing of micellar proteins. Lowerpart: reconstitution step. (c) Surface of the incubation well observedby RLM after detergent removal from a. Reconstituted membranescorrespond to the light gray background with defects appearing asdark gray circles. Bar: 50 Wm. (d) Low magni¢cation electronmicrograph of the membrane sheets formed at the interface after de-tergent removal. The large membrane is a single bilayer as shownby the folded sheet. Bar: 2 Wm.

FEBS 25156 23-8-01

D. Levy et al./FEBS Letters 504 (2001) 187^193 189

protein micelles were continuously transformed into very largemembrane sheets during detergent removal. The large recon-stituted membranes were several tens of Wm in size, coveringnearly all the electron microscope grid (Figs. 2c and 3c). Edgeviews, observed when lipid layer folded away from the grid,demonstrated that this continuous layer consisted of a singlebilayer (Fig. 2d). In addition, observations at high magni¢ca-tion clearly indicated the proteins in a densely packed or nearcrystalline arrangement in the membranes, denoting the e¤-ciency of the reconstitution process (Fig. 3c). Finally, themorphology of the membranes reconstituted at the surfacewas always large planar sheets but never vesicles or tubes asoften observed in 2D crystallization trials in volume.

3.3. 2D crystallizationAfter binding of the ternary lipid^detergent^protein mi-

celles and complete detergent removal, proteins make contactsin the reconstituted membrane and, in favorable cases, cancrystallize as observed for FhuA, TF0F1, bacteriorhodopsinand Anc2.

Besides the nature of the protein, critical parameters forsuccessful crystal formation were, as for the 2D crystallizationin volume, the lipid to protein ratio and the nature of the lipidadded to the micellar protein preparation before injection inthe crystallization well. Noteworthy, lipid to protein ratioshigher than those used in the volume method were neededto obtain large reconstituted membranes at the surface. In-deed, while reconstitution in volume involved both lipid^de-tergent and lipid^protein^detergent micelles, only the latter

were bound to the functionalized lipidic surface and partici-pated in the reconstitution step. Other parameters that haveto be adjusted experimentally for successful crystallization arerelated to the temperature, viscosity and bu¡er composition.

Concerning the quality of the 2D crystals obtained so far, ithas to be stressed that the lipid layer method generally pro-duced very large crystalline areas, much larger than thosegenerally produced by the conventional bulk method. How-ever, highly coherent areas in the reconstituted crystalline do-mains were too small to extract high resolution structuralinformation and resolutions of the best crystal extend toabout 10 Aî . One interpretation of this drawback could berelated to the major problem encountered with the the lipidlayer method, that is the di¤culty in the transfer of interfacial¢lms which can alter the crystallinity of the structures formedin situ. However, since several large coherent 2D crystals havebeen obtained with soluble proteins, the small size of highlycoherent crystals of membrane proteins cannot be related tounfavorable properties of the interface and its transfer. It isthus clear that for further improvements of the method, crys-tallogenesis studies are needed to understand the restrictedgrowth of coherent membrane protein 2D crystals formed atthe surface. In particular, a limiting factor could be a limitedfreedom for proteins to establish crystalline contacts in a re-constituted membrane once bound to a lipid ligand at thesurface.

Concerning the organization of the proteins in the 2D crys-tals, all 2D crystals obtained with the lipid layer method haveshown proteins incorporated in the same orientation. Suchspeci¢city is the result of the unidirectional orientation im-posed by the speci¢c binding of the protein to its lipid ligand.This has been clearly demonstrated in the case of FhuA, aprotein for which 2D crystals have been produced through theuse of the two strategies : while the classical 2D crystallizationin volume led to 2D crystals with proteins facing up and downin the plane of the membrane [36], the lipid layer led to 2Dcrystals with all FhuA molecules in the same orientation (Fig.4). This result also pointed out that 2D crystals obtained atthe surface could not be related to 2D crystals formed in thevolume of the droplet and then absorbed and fused to thelipid layer. Finally, the asymmetric protein orientation im-posed during the binding step has been demonstrated to bean advantage in 2D crystallization trials of TF0F1, a proteinwhich organizes in up and down orientations using crystalli-zation trials in volume. Although of low resolution, 2D crys-tals of TF0F1 produced with the lipid layer strategy representthe best 2D crystals obtained so far for the whole enzyme [21].It is likely that hydrophilic contacts through adjacent F1parts, which are the sole protein/protein interactions thatmay occur in an asymmetric orientation, are more favorablefor crystallization than hydrophilic and additional hydropho-bic interactions that may occur in the up and down orienta-tions.

4. Lipid ligands

Interfacial lipid ¢lms used for growing protein 2D crystalscontain a speci¢c lipid ligand and a second lipid, referred to asdilution lipid made of an unsaturated phospholipid to ensurea lipid layer in a £uid state. Mixtures of ligand lipid/dilutionlipid ranging from 0.75 to 0.25 (mol/mol) are frequentlyused.

Fig. 3. 2D crystallization of His-TF0F1 on Ni-NTA lipid layer. Ex-perimental conditions as in [21]. Upper part: binding step. (a) Sur-face of the incubation well observed by RLM showing the proteinmicellar domains formed after protein binding (light gray) and un-bound areas (*). Bar: 50 Wm. (b) High magni¢cation electron micro-graph of the transferred micellar domains. Bar: 200 nm. Lowerpart: reconstitution step. (c) Surface of the incubation well observedby RLM after detergent removal from a. Reconstituted membranescorrespond to the light gray background with defects appearing asdark gray breaks. Bar: 50 Wm. (d) High magni¢cation electron mi-crographs of the membrane sheets formed at the interface andshowing crystalline areas of the reconstituted proteins. Bar: 100 nm.

FEBS 25156 23-8-01

D. Levy et al./FEBS Letters 504 (2001) 187^193190

The protein to be crystallized may interact with the lipid vianon-speci¢c electrostatic interactions or via speci¢c interac-tions mediated by high-a¤nity ligands attached to the polarhead group of the lipids. To date, most of the experimentsreported for 2D crystallization of membrane proteins haveused a lipid derivatized with a Ni2-NTA head group mole-cule to link genetically modi¢ed recombinant proteins con-taining an accessible stretch of histidine (His) residues. Thislipid ligand already proven successful for 2D crystallization ofseveral His-tagged soluble proteins [37^41] has also been dem-onstrated e¤cient for membrane proteins [21,28]. Bu¡er con-ditions for the binding of His-tagged proteins to Ni2-NTAlipids have been reported similar to those reported for bindingof proteins to Ni2-NTA resins. In particular, the speci¢cbinding of His-tagged protein to the Ni2-chelating lipid re-quired low imidazole concentrations, the absence of EDTAand pH above 6.0. Di¡erences in the kinetics of the bindingof proteins to Ni-NTA lipids were observed, likely due todi¡erent accessibility of the His-tag. For example, bindingoccurred in less than 1 h at 20³C for proteins with a His-tag attached to a large external loop, as for (His)10 TF0F1or (His)6 FhuA, while overnight incubation at 4³C was neededfor binding more hydrophobic proteins with a His-tag at-tached to a short C-terminus, as for (His)6 melibiose permeaseor (His)6 Anc2. For protein concentrations below 100 nM (10Wg/ml for a 100 kDa protein) no binding was observed, inagreement with the values of dissociation constants between70 and 200 nM measured on immobilized Ni2-NTA surface[42].

More recently, binding and 2D crystallization of membraneproteins have been extended to non-speci¢c electrostatic inter-action using charged lipids as lipid ligands. For example,solubilized bacteriorhodopsin has been crystallized after bind-ing to positively charged lipid layer leading to 2D crystalsdi¡racting to 10 Aî resolution in ice (Fig. 5a). On the otherhand, 2D crystals of the mitochondrial transporter, Anc2,have been produced after binding to negatively charged sur-face (Fig. 5b), with even better reconstitution than that ob-tained with a Ni-chelating surface, likely due to the low ac-cessibility of the His-tag of this protein. Thus, electrostaticbinding of a protein to positively or negatively charged lipidscan be adapted depending upon the charges distribution ofthe membrane proteins and opens the ¢eld of lipid layer crys-

tallization to those membrane proteins that cannot be genet-ically modi¢ed.

5. Experimental set-up and re£ected light microscopy (RLM)

Experiments are done in 60 Wl te£on wells, with a 4 mmdiameter, i.e. the same size as the electron microscopy carbon-coated grids which are deposited on the crystallization dropfor the transfer of the structures formed at the interface. Onthe side of each well, an injection hole allows addition ofmicellar protein solutions or Bio-Beads under the lipid layer.

To follow the binding to the surface and the 2D crystalli-zation steps, several techniques have been used in the case ofsoluble proteins, including epi£uorescence, ellipsometry, sur-face pressure measurement, viscosimetry, Brewster angle mi-croscopy and electron microscopy [43^46]. However, theseapproaches require large wells and are material consuming,precluding their use in 2D crystallization trials of membraneproteins. We have found that RLM was a powerful tool tofollow the important steps of protein binding and reconstitu-tion at the interface. Furthermore, observations are performeddirectly on the surface of the crystallization wells and allow afast screening in situ of the crystallization parameters (Chamiet al., in preparation).

After spreading the lipid layer on the detergent-free bu¡ersolution, the surface focused by RLM appeared as a homoge-neous gray background except for some small aggregates oflipid £oating at the surface. The easily observed movement ofthese lipid aggregates is in agreement with the £uid state ofthe spread lipid layer. Following injection of a micellar pro-tein solution, signi¢cant changes of the surface can be ob-served: (i) the surface £uidity decreased and appeared frozenwith no more movement; (ii) the surface is progressively cov-ered by whiter areas (Figs. 2a and 3a) which increased in size,covering nearly all the surface of the well. Control experi-ments with His-tagged proteins demonstrated that these opti-cal changes were correlated to the speci¢c binding to Ni2-NTA lipids since they were not observed in the presence ofhigh imidazole concentrations or in the presence of EDTA. Inaddition, after transfer of the surface to carbon-coated grids,the white areas observed in RLM corresponded clearly to theprotein-containing planar domains observed by electron mi-croscopy (Figs. 2b and 3b).

Fig. 4. Protein orientation in di¡erent 2D crystals of FhuA. (a) Average projection map from 2D crystals produced by the lipid layer method[21]. Inset: re£ections correspond to IQ 1 (b) and IQ 2^4 (a) and rings to 1/40, 1/20, 1/15 Aî 31. (b) Average projection map from 2D crystalsproduced by the method of 2D crystallization in volume [36]. Projection maps calculated at 15 Aî resolution. FhuA dimers are delineated ineach map by an S line. The schemes indicate the orientations of FhuA dimers in both 2D crystals.

FEBS 25156 23-8-01

D. Levy et al./FEBS Letters 504 (2001) 187^193 191

Upon detergent removal, the contrast of the surface focusedby RLM increased drastically, concomitant with the transfor-mation of the micellar layer into reconstituted bilayer. Recon-stituted membrane appeared by RLM as large light gray areascovering all the crystallization well, i.e. 4 mm diameter (Figs.2c and 3c). Interestingly, again there was a good correlationbetween the morphology and the size of the structures ob-served by RLM and the structures observed by electron mi-

croscopy after transfer of the lipid layer. Thus, despite RLMprovides information on a microscopic range, it represents apowerful tool to visualize in situ two important steps of thecrystallization process, i.e. binding and reconstitution by de-tergent removal. The kinetics of both processes can be con-trolled through the structures observed by optical microscopyand some parameters of the reconstitution adjusted withoutthe need for time consuming electron microscopy observa-tions.

6. Conclusions and perspectives

Membrane proteins have been poorly characterized com-pared to soluble proteins, because they are embedded in alipidic environment and thus require special treatment to bestudied. This hindrance is clearly illustrated by the few num-ber of high resolution structural data on membrane proteins,itself related to the di¤culty in crystallizing these membraneproteins. Thus, a considerable interest exists for designing in-novative strategies to produce 2D or 3D crystals of membraneproteins that are amenable to structural analysis by electronor X-ray crystallography.

This manuscript reports on the lipid layer 2D crystallizationstrategy which can be foreseen as a new promising alternativeto the conventional method for membrane protein 2D crys-tallization. This strategy shown recently e¤cient for the crys-tallization of a very diverse range of membrane proteins cannow be run in parallel to the classical method in volume andwill increase the chances of success. Further detailed studies ofcrystallogenesis are currently in progress to improve the strat-egy in terms of producing highly coherent 2D crystals and todemonstrate its general applicability, using other membraneproteins solubilized in di¡erent classes of detergent.

Although, as other crystallization methods, the lipid layermethod has its own disadvantages related to reproducibility,quality of the transfer and size of highly coherent 2D crystals,it gains the important advantage to require very smallamounts of proteins per crystallization trials. Thus this meth-od opens the ¢eld of 2D crystallization to membrane proteinswhich are produced and puri¢ed in low amounts and will beof particular help for studying receptors, channels and othermembrane eukaryotic proteins which are, to date, di¤cult tooverexpress. In addition, although the vast applicability of theNTA lipid already provides a common moiety for the bindingof many diverse His-tagged membrane proteins, the possibilityto extend the lipid layer method to other speci¢c a¤nity bind-ing is expected to make this strategy even more general. Forexample, the use of non-speci¢c electrostatic interactions be-tween membrane proteins and charged lipids has already beendemonstrated e¤cient. Interestingly, several functionalizedlipids already developed to bind soluble proteins could alsobe used such as lipids with a head group made of ATP ana-logues to bind proteins with accessible nucleotide binding sites[47], biotin group for avidin genetically modi¢ed proteins [48]or N-ethylmaleimide groups for cysteine-containing proteins[49].

Finally, further development of the lipid layer strategy forproviding structural information on membrane proteinswould be: (1) to take advantage of the £atness of the lipidsurface which provides the formation of large and planarreconstituted membranes. Such reconstituted membranes inwhich proteins are incorporated in a preferred orientation

Fig. 5. 2D crystallization on charged lipidic surface. (a) Electron mi-crograph of the 2D crystals of bacteriorhodopsin produced on posi-tively charged lipid layer. A mixture of delipidated bacteriorhodop-sin solubilized in dodecylmaltoside was supplemented with eggphosphatidylcholine, injected and reconstituted at 25 Wg/ml beneatha lipid layer made of 1,2-dioleoyl-3-dimethyl-ammonium propane.Bar: 5 Wm. Inset: optical di¡raction pattern with spots up to 10 Aîin ice (a). (b) Electron micrograph of the 2D crystals of Anc2 pro-duced on negatively charged lipid layer. A mixture of His Anc2solubilized in dodecylmaltoside was supplemented with egg phospha-tidylcholine, injected and reconstituted at 30 Wg/ml beneath a lipidlayer made of egg phosphatidic acid. Bar: 2 Wm. Inset: optical dif-fraction pattern with spots up to 18 Aî in ice (a).

FEBS 25156 23-8-01

D. Levy et al./FEBS Letters 504 (2001) 187^193192

and at high density would be a suitable material for highresolution atomic force microscopy [50]; (2) to use the lipidlayer as a template for fusion of pre-formed small 2D crystals[51]. Preliminary experiments have shown the feasibility ofthis approach by fusing, on a positively charged lipid layer,small vesicular 2D crystals of the mechanosensitive channelMscl produced by detergent removal in volume; (3) the use offunctionalized lipid layers to concentrate protein micelles atan interface and that can serve as seeds for epithaxial growthin membrane protein 3D crystallization trials, as it has beendemonstrated in the case of soluble proteins [52].

References

[1] Jones, D.T. (1998) FEBS Lett. 43, 281^285.[2] Grisshammer, R. and Tate, C.G. (1995) Q. Rev. Biophys. 28,

315^422.[3] Kiefer, H., Krieger, J., Olszewski, J.D., Von Heijne, G., Prest-

wich, G.D. and Breer, H. (1996) Biochemistry 35, 16077^16084.[4] Miroux, B. and Walker, J.E. (1996) J. Mol. Biol. 260, 289^298.[5] Ostermeier, C. and Michel, H. (1997) Curr. Opin. Struct. Biol. 7,

697^701.[6] Iwata, S., Ostermeier, C., Ludwig, B. and Michel, H. (1995)

Nature 376, 660^665.[7] Landau, E.M. and Rosenbusch, J.P. (1996) Proc. Natl. Acad. Sci.

USA 93, 14532^14535.[8] Rummel, G., Hardmeyer, A., Widner, C., Chiu, M.L., Nollert,

P., Pedruzzi, I., Landau, E.M. and Rosenbusch, J.P. (1998)J. Struct. Biol. 121, 82^91.

[9] Henderson, R., Baldwin, J.M., Ceska, T.A., Zemlin, F., Beck-mann, E. and Downing, K.H. (1990) J. Mol. Biol. 213, 899^929.

[10] Ku«hlbrandt, W., Wang, D.N. and Fujiyoshi, Y. (1994) Nature367, 614^621.

[11] Murata, K., Mitsuoka, K., Hirai, T., Walz, T., Agre, P., Hey-mann, J.B., Engel, A. and Fujiyoshi, Y. (2000) Nature 407, 599^605.

[12] Stahlberg, H., Braun, T., de Groot, B., Philipsen, A., Borgnia,M., Agre, P., Kuhlbrandt, W. and Engel, A. (2000) J. Struct.Biol. 132, 133^142.

[13] Stokes, D.L., Auer, M., Zhang, P. and Kuhlbrandt, W. (1999)Curr. Biol. 9, 672^679.

[14] Rigaud, J.L., Pitard, B. and Lëvy, D. (1995) Biochim. Biophys.Acta 1231, 223^246.

[15] Ku«hlbrandt, W. (1992) Q. Rev. Biophys. 25, 1^49.[16] Jap, B.K., Zulauf, M., Scheybani, T., Hefti, A., Baumeister, W.,

Aebi, U. and Engel, A. (1992) Ultramicroscopy 46, 45^84.[17] Heymann, J.B., Muller, D.J., Mitsuoka, K. and Engel, A. (1997)

Curr. Opin. Struct. Biol. 7, 543^549.[18] Hasler, L., Heymann, J.B., Engel, A., Kistler, J. and Walz, T.

(1998) J. Struct. Biol. 121, 162^171.[19] Walz, T. and Grigorie¡, N. (1998) J. Struct. Biol. 121, 142^161.[20] Rigaud, J.L., Chami, M., lambert, O., Lëvy, D. and Ranck, J.L.

(2000) Biochim. Biophys. Acta 1508, 112^128.[21] Lëvy, D., Mosser, G., Lambert, O., Moeck, G., Bald, D. and

Rigaud, J.L. (1999) J. Struct. Biol. 127, 44^52.[22] Kornberg, R.D. and Darst, S. (1991) Curr. Opin. Struct. Biol. 1,

642^646.[23] Brisson, A., Olofsson, A., Ringler, P., Schmutz, M. and Stoylova,

S. (1994) Biol. Cell 80, 221^228.

[24] Chiu, W., Avila-Sakar, A.J. and Schmid, M.F. (1997) Adv. Bio-phys. 34, 161^172.

[25] Asturias, F.J. and Kornberg, R.D. (1999) J. Biol. Chem. 274,6813^6816.

[26] Leuther, K.K., Bushnell, D.A. and Kornberg, R.D. (1996) Cell85, 773^779.

[27] Avila-Sakar, A.J. and Chiu, W. (1996) Biophys. J. 70, 57^68.[28] Lebeau, L., Lasch, F., Vënien-Bryan, C., Renault, A., Dietrich,

J., Jahn, T., Palmgren, M., Kuªhlbrandt, W. and Mioskowski, C.(2001) J. Mol. Biol. 308, 639^647.

[29] Lëvy, D., Gulick, A., Seigneuret, M. and Rigaud, J.L. (1990)Biochemistry 29, 9480^9488.

[30] Paternostre, M., Roux, M. and Rigaud, J.L. (1988) Biochemistry27, 2666^2676.

[31] Vinson, P.K., Talmon, Y. and Walter, A. (1989) Biophys. J. 56,669^681.

[32] Almog, S., Litman, B.J., Winley, W., Cohen, J., Wachtel, E.J.,Barenholz, Y., Ben-Shaul, A. and Lichyenberg, D. (1990) Bio-chemistry 29, 4582^4592.

[33] Lambert, O., Lëvy, D., Ranck, J.L., Leblanc, G. and Rigaud,J.L. (1998) Biophys. J. 74, 918^930.

[34] Rigaud, J.L., Lëvy, D., Mosser, G. and Lambert, O. (1998) Eur.J. Biophys. 27, 305^319.

[35] Rigaud, J.L., Mosser, G., Lacapëre, J.J., Olofsson, A., Lëvy, D.and Ranck, J.L. (1997) J. Struct. Biol 118, 226^235.

[36] Lambert, O., Moeck, G., Levy, D., Planc°on, L., Letellier, L. andRigaud, J.L. (1999) J. Struct. Biol. 126, 145^155.

[37] Kubalek, E.W., Le Grice, S.F. and Brown, P.O. (1994) J. Struct.Biol. 113, 117^123.

[38] Barklis, E., McDermott, J., Wilkens, S., Schabtach, E., Schmid,M.F., Fuller, S., Karanjia, S., Love, Z., Jones, R., Rui, Y., Zhao,X. and Thompson, D. (1997) EMBO J. 16, 1199^1213.

[39] Venien-Bryan, C., Balavoine, F., Toussaint, B., Mioskowski, C.,Hewat, E.A., Helme, B. and Vignais, P.M. (1997) J. Mol. Biol.274, 687^692.

[40] Wilson-Kubalek, E.M., Brown, R.E., Celia, H. and Milligan,R.A. (1998) Proc. Natl. Acad. Sci. USA 95, 8040^8045.

[41] Celia, H., Wilson-Kubalek, E., Milligan, R.A. and Teyton, L.(1999) Proc. Natl. Acad. Sci. USA 96, 5634^5639.

[42] Kroger, D., Liley, M., Schiweck, W., Skerra, A. and Vogel, H.(1999) Biosens. Bioelectron. 14, 155^161.

[43] Shneck, D.R., Pack, D.W., Sasaki, D.Y. and Arnold, F.H. (1994)Langmuir 10, 2382^2388.

[44] Frey, W., Schief Jr., W.R., Pack, D.W., Chen, C.T., Chilkoti, A.,Stayton, P., Vogel, V. and Arnold, F.H. (1996) Proc. Natl. Acad.Sci. USA 93, 4937^4941.

[45] Venien-Bryan, C., Lenne, P.F., Zakri, C., Renault, A., Brisson,A., Legrand, J.F. and Berge, B. (1998) Biophys. J. 74, 2649^2657.

[46] Dorn, I.T., Neumaier, K.R. and Tampe, R. (1998) J. Am. Chem.Soc. 12, 2753^2763.

[47] Aribi, H.O., Reichart, P. and Kornberg, R.D. (1987) Biochemis-try 26, 7974^7980.

[48] Liu, Z., Qin, C., Xiao, S., Wang, W. and Sui, S.F. (1995) Eur.Biophys. J. 24, 31^37.

[49] Brink, G., Schmitt, L., Tampë, R. and Sackman, E. (1994) Bio-chim. Biophys. Acta 1196, 227^230.

[50] Scheuring, S., Muller, D., Ringler, P., Hetmann, J.B. and Engel,A. (1999) J. Microsc. 193, 28^35.

[51] Gritsch, S., Neumeier, K., Schmitt, L. and Tampe, R. (1995)Biosens. Bioelectron. 10, 805^812.

[52] Edwards, A.M., Darst, S.A., Hemming, S.A., Li, Y. and Korn-berg, R.D. (1994) Nat. Struct. Biol. 1, 195^197.

FEBS 25156 23-8-01

D. Levy et al./FEBS Letters 504 (2001) 187^193 193